Pith. sign in

REVIEW 3 major objections 5 minor 79 references

Incorporating Visual Correspondence into Diffusion Model for Virtual Try-On

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Explicitly matching garment points to body locations in the diffusion denoiser preserves fine texture and shape in virtual try-on, the paper argues.

desk verdict A genuinely useful, well-ablated VTON method paper whose central claim survives the circularity concern; the real blemishes are tune-on-test hyperparameters and the missing GarDiff comparison. read the letter →

arxiv 2505.16977 v1 pith:CZROIWNL submitted 2025-05-22 cs.CV cs.MM

Siqi Wan , Jingwen Chen , Yingwei Pan , Ting Yao , Tao Mei This is my paper · ORCID
classification cs.CVcs.MM
keywords virtualtry-ondiffusionmodelvisualcorrespondencesemanticpointmatchingpoint-focusedlossgarmentdetailpreservation3D-awarefeatureaugmentationdual-branchUNet
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Diffusion-based virtual try-on typically feeds the whole in-shop garment into a second UNet and hopes the denoiser learns where each detail belongs on the body; the paper's diagnosis is that this implicit route lets the stochastic denoising process blur or invent texture. SPM-Diff instead treats try-on as a point-matching problem: it samples interest points on the garment, warps them onto the target person with a local flow field, augments the 2D cues with depth and normal maps, and injects the matched features into selected self-attention layers of the main denoising UNet. A point-focused loss then places extra reconstruction weight exactly at those matched locations. On VITON-HD and DressCode, the method reports state-of-the-art or near-state-of-the-art scores and user-study preferences for garment detail, shape, and realism. The payoff is that an explicit correspondence prior tames diffusion randomness, so fine-grained garment details survive the sampling process.

What carries the argument

Semantic point matching (SPM): a set of interest points sampled on the in-shop garment, mapped to the target person through a learned dense displacement field, augmented with the target person's depth/normal features, and injected into the self-attention layers of the Main-UNet. The companion point-focused loss (Eq. 8) makes the denoiser reconstruct those exact locations with extra weight, converting the correspondence prior into a training signal.

What would settle it

On a held-out set of garment–person pairs with manually annotated semantic-point correspondences, replace the GP-VTON-style flow with a deliberately corrupted or identity displacement field; if SPM-Diff with the point-focused loss then performs no better than the Base+Geo map ablation across SSIM, LPIPS, FID, and KID, the claim that visual correspondence is the causal mechanism would be falsified.

Watch

Extended reading notes

Core claim

SPM-Diff reframes virtual try-on as semantic point matching. The authors sample $K$ interest points $P_G$ on the in-shop garment using SuperPoint and farthest-point sampling, then project them to the target person with a GP-VTON-style local-flow global-parsing warping module, giving $P_H = P_G + M_{G\to H}$ via a dense displacement field. Features at those points are extracted, the garment-point features are augmented into 3D-aware cues by adding features from the target person's depth and normal maps, and the results are injected into the first two and last two self-attention layers of the Main-UNet. Training uses a point-focused diffusion loss $\|\epsilon - \hat{\epsilon}\|_2^2 + \lambda\|(\epsilon-\hat{\epsilon})\odot M(P_H)\|_2^2$, which upweights noise-prediction error exactly at the warped semantic points. The reported result is consistent gains in SSIM, LPIPS, FID, and KID against GAN-based and diffusion-based baselines, with the SPM component alone giving the largest relative jumps in the ablation.

Load-bearing premise

The paper's gains rest on the local-flow warping module producing a displacement field accurate enough that each sampled garment point maps to its true corresponding spot on the body; if the flow is wrong, the point-focused loss teaches the model to put garment details at the wrong locations.

Editorial extensions

If this is right

  • Any dual-branch diffusion try-on model could adopt SPM as a modular addition, since it only changes where and how features are injected and adds one loss term.
  • The point-focused loss biases denoising toward fidelity at the matched locations, which should reduce the number of sampling steps needed for detail preservation, although the paper keeps the standard 20-step UniPC sampler.
  • Because the injection happens in latent space rather than on the pixel-level warped garment, the method tolerates imperfect flow, which the paper argues makes it robust to challenging poses and loose garments.
  • The gains transfer across VITON-HD, DressCode categories, cross-dataset evaluation, and a stronger SDXL backbone, suggesting the correspondence prior is not dataset-specific.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An untested extension is to apply the same semantic-point mask at inference as attention re-weighting rather than only as a training loss, which would let users emphasize detail locations without retraining.
  • The same recipe—sample semantic points, warp them to a target geometry, and upweight reconstruction at those points—could transfer to other conditional diffusion tasks where precise alignment matters, such as object insertion, face swapping, or image compositing.
  • The paper fixes K=25 and reports degradation at K=50/75 due to flow imprecision; a confidence-weighted point selection based on the warping module's local flow error would likely extend the usable range, a direction the authors do not explore.
  • The reported point-matching MSE of 29.84 for the warping module versus 39.46 for DIFT suggests flow accuracy is the current ceiling; if the ceiling were raised by a better correspondence estimator, SPM-Diff's gains should grow proportionally.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes SPM-Diff, a dual-branch latent diffusion model for virtual try-on that explicitly injects visual correspondence between the in-shop garment and the target person. Semantic points on the garment are detected with SuperPoint, warped to the person with a GP-VTON-style local flow module (Eq. 4), augmented with depth/normal-map features, and injected into the self-attention layers of the Main-UNet (Eqs. 5-7). A point-focused diffusion loss (Eq. 8) upweights the reconstruction error at the warped points. Experiments on VITON-HD and DressCode show improvements over several GAN- and diffusion-based baselines, with ablations for the SPM module, the point count K, the loss weight λ, and cross-dataset evaluation.

Significance. If the correspondence mechanism is sound, the idea of using sparse semantic points to tame diffusion stochasticity is a useful and potentially generalizable contribution to virtual try-on and conditional image generation. The paper provides extensive ablations, consistent improvements over strong baselines, a fair comparison with IDM-VTON_sdxl using the same SDXL base, and public code. However, the central claim that the gains come from accurate visual correspondence is weakened by the reliance on an imperfect estimated flow and by a circular evaluation metric; the significance of the contribution therefore depends on whether these issues are resolved with additional validation.

major comments (3)
  1. [Sec. 3.2.2, Eq. (4) and Eq. (8)] The point-focused loss and the SPM feature injection both use P_H computed from the same estimated flow M_{G->H}. There is no independent supervision on correspondence accuracy, so the training signal can be satisfied by reproducing the warper's errors rather than by learning true garment-person correspondence. The paper's own ablation in Sec. 4.3 acknowledges that M_{G->H} is imprecise ("significant offsets from imprecise local flow map M_{G->H}"). To support the central claim, the authors should test robustness to flow errors, for example by training with ground-truth or manually annotated correspondences on a subset, by replacing the warper with an oracle or a different warping method, and by reporting how performance degrades when the flow is intentionally perturbed.
  2. [Sec. 4.3, Effect of SPM in Visual Correspondence] The correspondence evaluation in Sec. 4.3 uses SuperPoint both for sampling P_G and for measuring matching between the input garment and the synthesized person image. This is circular: it shows that the output contains SuperPoint-detectable keypoints, not that those keypoints correspond semantically to the garment. The metric should be based on manually annotated semantic correspondences (e.g., PCK on annotated pairs) or a separate correspondence benchmark, rather than the same detector used to define the training signal.
  3. [Appendix, Tables 5-7 and Table 6] The hyperparameters K and λ are selected on the VITON-HD benchmark, and the same benchmark is used for the final numbers in Tables 1-3. This selection-on-test-set procedure may inflate the reported gains. The authors should either tune on a separate validation split or clearly report the sensitivity of results on DressCode and the cross-dataset settings with the selected values, explaining how the selection generalizes.
minor comments (5)
  1. [Abstract and Table 1] The abstract and introduction claim "state-of-the-art" performance, but Table 1 shows that SPM-Diff (SD1.5) is not uniformly better than IDM-VTON_sdxl on VITON-HD; SPM-Diff_sdxl is only marginally better. The claim should be softened to "on par with or better than" for each metric.
  2. [Eq. (5)] The "3D-aware" augmentation is implemented as a simple addition of the geometry feature F_H to the garment feature F_G. The authors should clarify how this operation goes beyond concatenation and why it is described as "3D-aware" beyond the fact that F_H is derived from depth/normal maps.
  3. [Sec. 3.2.2 and Implementation Details] The caption of Fig. 2 mentions "SMPL model" while the implementation section states the depth/normal maps are estimated with OSX. These references should be made consistent.
  4. [Appendix A.4] The flow warping module is said to follow GP-VTON, but no training details are provided for this module, such as whether it is frozen or jointly fine-tuned, what loss is used, and how it is initialized. These details are needed for reproducibility.
  5. [Table 3] For the VITON-HD/SSHQ-1.0 evaluation only FID and KID are reported. Reporting LPIPS and SSIM, even if the setting is unpaired, would allow a more complete comparison with the other cross-dataset rows.

Circularity Check

0 steps flagged · score 2.0 of 10

No construction-level circularity: the SPM prior is an explicit warped-point input and benchmark gains are external; only minor related-work self-citations and a mild SuperPoint-based evaluation overlap keep the score above zero.

full rationale

SPM-Diff's derivation is self-contained at the construction level. The warped point set PH is an explicit algorithmic input computed by a GP-VTON-style flow module (external, Appendix A.4), not a quantity the diffusion model predicts; Eq. 8 merely up-weights the standard noise-reconstruction loss at those coordinates. Since the final image is trained against ground-truth noise and scored on held-out SSIM/LPIPS/FID/KID (Tables 1-3) plus ablations (Table 4), there is no fitted parameter renamed as a prediction. The same PH is used in feature injection and in the loss mask, which means flow errors are shared, but the paper acknowledges this sensitivity (Sec 4.3, K=50/75) and validates the flow against manually annotated points (Table 9), so this is a stated assumption, not a circular derivation. The only mild overlaps are (i) related-work self-citations (e.g., GarDiff, Wan et al. 2024) that are not load-bearing, and (ii) the Fig. 5 correspondence metric uses SuperPoint, the same detector used to sample semantic points, so that particular metric is not independent; however, it is a supporting visualization, not the basis of the SOTA claim. No uniqueness theorem or self-citation chain is used to force the design. Score 2 reflects these minor overlaps, not construction-level circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central result rests on two tuned hyperparameters (λ and K), on the reliability of the flow warping and the OSX 3D reconstruction, on SuperPoint's ability to select detail-relevant points, and on the pretrained Stable Diffusion/VAE prior. No new physical or model entities are postulated; semantic points are operationalized with existing detectors and warping models.

free parameters (2)
  • λ (point-focused loss weight) = 0.5
    Selected on VITON-HD by sweeping {1.0, 0.5, 0.1, 0.05, 0.01} (Table 6); affects all final reported results.
  • K (semantic point count) = 25
    Selected on VITON-HD by sweeping {10, 25, 50, 75} (Table 5) with a sensitivity check around 20 to 40 (Table 7); K=25 gives best SSIM/LPIPS/FID/KID.
assumptions (4)
  • domain assumption Pretrained Stable Diffusion (SD 1.5 / SDXL) and its VAE provide a strong generative prior suitable for VTON fine-tuning.
    Sec. 3.1 and Sec. 4.1: the model is initialized from SD and inherits its image prior; the method's absolute performance depends on this upstream model.
  • domain assumption The GP-VTON-style flow warping module yields a reliable dense displacement field M_{G→H} between garment and person.
    Eq. 4 in Sec. 3.2.2; the SPM feature injection and point-focused loss are applied at flow-warped locations; the paper's K ablation shows flow imprecision degrades results.
  • domain assumption OSX (Lin et al. 2023) provides garment-agnostic depth/normal maps that correctly represent the target person's 3D body.
    Sec. 3.2.2 and Appendix A.4; the 3D-aware feature augmentation (Eq. 5) fuses point features with geometry features extracted from these maps.
  • domain assumption SuperPoint interest points plus farthest point sampling select points that represent the garment's fine-grained appearance and shape.
    Sec. 3.2.2 Point Sampling; the method assumes sparse detector-based points capture the details that need preservation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Incorporating Visual Correspondence into Diffusion Model for Virtual Try-On." pith.science (2026). https://pith.science/paper/CZROIWNL

@misc{pith2026250516977,
  author       = {Pith},
  title        = {Pith review of: Incorporating Visual Correspondence into Diffusion Model for Virtual Try-On},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CZROIWNL}},
  note         = {Machine review of arXiv:2505.16977}
}
read the original abstract

Diffusion models have shown preliminary success in virtual try-on (VTON) task. The typical dual-branch architecture comprises two UNets for implicit garment deformation and synthesized image generation respectively, and has emerged as the recipe for VTON task. Nevertheless, the problem remains challenging to preserve the shape and every detail of the given garment due to the intrinsic stochasticity of diffusion model. To alleviate this issue, we novelly propose to explicitly capitalize on visual correspondence as the prior to tame diffusion process instead of simply feeding the whole garment into UNet as the appearance reference. Specifically, we interpret the fine-grained appearance and texture details as a set of structured semantic points, and match the semantic points rooted in garment to the ones over target person through local flow warping. Such 2D points are then augmented into 3D-aware cues with depth/normal map of target person. The correspondence mimics the way of putting clothing on human body and the 3D-aware cues act as semantic point matching to supervise diffusion model training. A point-focused diffusion loss is further devised to fully take the advantage of semantic point matching. Extensive experiments demonstrate strong garment detail preservation of our approach, evidenced by state-of-the-art VTON performances on both VITON-HD and DressCode datasets. Code is publicly available at: https://github.com/HiDream-ai/SPM-Diff.

Figures

Figures reproduced from arXiv: 2505.16977 by the authors.

Figure 1
Figure 1. Illustration of given target person and (a) in-shape garment with semantic points. Existing [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall framework of our SPM-Diff. (a) Illustration of our semantic point matching [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Qualitative results on the VITON-HD dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (13 more)
Figure 5
Figure 5. Figure 5: Accuracy of visual correspondence between in-shop garment and synthesized person image. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: More qualitative results on the VITON-HD dataset. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results on the DressCode dataset. [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Ablation study of the pivotal components in SPM-Diff on VITON-HD dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Ablation study of semantic point count K on VITON-HD dataset. Evaluation of 3D Conditions Learnt by Various Methods. The 3D cues (i.e., depth/normal map) adopted in our training and inference are the estimated results by using pre-trained 3D human reconstruction model …
Figure 10
Figure 10. Figure 10: Tryon results on in-the-wild person images with complex backgrounds. [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Comparisons with warp-based methods on the VITON-HD dataset. [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: The visualization results of the point matching. [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Tryon result for loose clothes like dresses or skirts. [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]
Figure 14
Figure 14. Figure 14: Tryon result of full-body outfits generated by our SPM-Diff. [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]
Figure 15
Figure 15. Figure 15: Overview of the flow warping module. C FGA FG Self -Attn Kg K Alignment Feature Cross -Attn Garment CLIP Feature Feature Augmen t (2d→3 d ) Kg: Garm-UNet Hidden State K: Main-UNet Hidden State C FGA FG Self -Attn Kg K Alignment Feature Cross -Attn Garment CLIP Feature…
Figure 16
Figure 16. Figure 16: Details of the feature injection process. [PITH_FULL_IMAGE:figures/full_fig_p021_16.png]
Figure 17
Figure 17. Figure 17: SPM-Diff suffers from artifacts in human hands and fails to fully preserve decorative [PITH_FULL_IMAGE:figures/full_fig_p022_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 54 canonical work pages

  1. [1]

    Blended diffusion for text-driven editing of natural images

    Omri Avrahami, Dani Lischinski, and Ohad Fried. Blended diffusion for text-driven editing of natural images. In CVPR, 2022

    work page 2022

  2. [2]

    Single stage virtual try-on via deformable attention flows

    Shuai Bai, Huiling Zhou, Zhikang Li, Chang Zhou, and Hongxia Yang. Single stage virtual try-on via deformable attention flows. In ECCV, 2022

    work page 2022

  3. [3]

    Demystifying mmd gans

    Miko aj Bi \'n kowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. In ICLR, 2018

    work page 2018

  4. [4]

    Bookstein

    Fred L. Bookstein. Principal warps: Thin-plate splines and the decomposition of deformations. IEEE TPAMI, 11 0 (6): 0 567--585, 1989

    work page 1989

  5. [5]

    Masactrl: Tuning-free mutual self-attention control for consistent image synthesis and editing

    Mingdeng Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xiaohu Qie, and Yinqiang Zheng. Masactrl: Tuning-free mutual self-attention control for consistent image synthesis and editing. In CVPR, 2023

    work page 2023

  6. [6]

    Toward realistic image compositing with adversarial learning

    Bor-Chun Chen and Andrew Kae. Toward realistic image compositing with adversarial learning. In CVPR, 2019

    work page 2019

  7. [7]

    Controlstyle: Text-driven stylized image generation using diffusion priors

    Jingwen Chen, Yingwei Pan, Ting Yao, and Tao Mei. Controlstyle: Text-driven stylized image generation using diffusion priors. In ACM MM, 2023 a

    work page 2023

  8. [8]

    Anydoor: Zero-shot object-level image customization

    Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. Anydoor: Zero-shot object-level image customization. In CVPR, 2024 a

    work page 2024

Show all 79 references
  1. [9]

    Control3d: Towards controllable text-to-3d generation

    Yang Chen, Yingwei Pan, Yehao Li, Ting Yao, and Tao Mei. Control3d: Towards controllable text-to-3d generation. In ACM MM, 2023 b

    2023

  2. [10]

    Improving text-guided object inpainting with semantic pre-inpainting

    Yifu Chen, Jingwen Chen, Yingwei Pan, Yehao Li, Ting Yao, Zhineng Chen, and Tao Mei. Improving text-guided object inpainting with semantic pre-inpainting. In ECCV, 2024 b

    2024

  3. [11]

    Viton-hd: High-resolution virtual try-on via misalignment-aware normalization

    Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In CVPR, 2021

    2021

  4. [12]

    Improving diffusion models for authentic virtual try-on in the wild

    Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for authentic virtual try-on in the wild. In ECCV, 2024

    2024

  5. [13]

    Zflow: Gated appearance flow-based virtual try-on with 3d priors

    Ayush Chopra, Rishabh Jain, Mayur Hemani, and Balaji Krishnamurthy. Zflow: Gated appearance flow-based virtual try-on with 3d priors. In ICCV, 2021

    2021

  6. [14]

    Dovenet: Deep image harmonization via domain verification

    Wenyan Cong, Jianfu Zhang, Li Niu, Liu Liu, Zhixin Ling, Weiyuan Li, and Liqing Zhang. Dovenet: Deep image harmonization via domain verification. In CVPR, 2020

    2020

  7. [15]

    Superpoint: Self-supervised interest point detection and description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superpoint: Self-supervised interest point detection and description. In CVPRW, 2018

    2018

  8. [16]

    Towards multi-pose guided virtual try-on network

    Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bochao Wang, Hanjiang Lai, Jia Zhu, Zhiting Hu, and Jian Yin. Towards multi-pose guided virtual try-on network. In CVPR, 2019 a

    2019

  9. [17]

    Fw-gan: Flow-navigated warping gan for video virtual try-on

    Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bowen Wu, Bing-Cheng Chen, and Jian Yin. Fw-gan: Flow-navigated warping gan for video virtual try-on. In ICCV, 2019 b

    2019

  10. [18]

    C-vton: Context-driven image-based virtual try-on network

    Benjamin Fele, Ajda Lampe, Peter Peer, and Vitomir Struc. C-vton: Context-driven image-based virtual try-on network. In WACV, 2022

    2022

  11. [19]

    Stylegan-human: A data-centric odyssey of human generation

    Jianglin Fu, Shikai Li, Yuming Jiang, Kwan-Yee Lin, Chen Qian, Chen Change Loy, Wayne Wu, and Ziwei Liu. Stylegan-human: A data-centric odyssey of human generation. In ECCV, 2022

    2022

  12. [20]

    Parser-free virtual try-on via distilling appearance flows

    Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In CVPR, 2021

    2021

  13. [21]

    Humans in 4d: Reconstructing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Humans in 4d: Reconstructing and tracking humans with transformers. In ICCV, 2023

    2023

  14. [22]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NeurIPS, 2014

    2014

  15. [23]

    Taming the power of diffusion models for high-quality virtual try-on with appearance flow

    Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. In ACM MM, 2023

    2023

  16. [24]

    Viton: An image-based virtual try-on network

    Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In CVPR, 2018

    2018

  17. [25]

    Clothflow: A flow-based model for clothed person generation

    Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. Clothflow: A flow-based model for clothed person generation. In ICCV, 2019

    2019

  18. [26]

    Style-based global appearance flow for virtual try-on

    Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In CVPR, 2022

    2022

  19. [27]

    Unsupervised semantic correspondence using stable diffusion

    Eric Hedlin, Gopal Sharma, Shweta Mahajan, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. Unsupervised semantic correspondence using stable diffusion. In NeurIPS, 2023

    2023

  20. [28]

    Prompt-to-prompt image editing with cross attention control

    Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Prompt-to-prompt image editing with cross attention control. In ICLR, 2023

    2023

  21. [29]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In NeurIPS, 2017

    2017

  22. [30]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022

    2022 arXiv

  23. [31]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In NeurIPS, 2020

    2020

  24. [32]

    Animate anyone: Consistent and controllable image-to-video synthesis for character animation

    Li Hu, Xin Gao, Peng Zhang, Ke Sun, Bang Zhang, and Liefeng Bo. Animate anyone: Consistent and controllable image-to-video synthesis for character animation. In CVPR, 2024

    2024

  25. [33]

    Openclip, July 2021

    Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, Cade Gordon, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, Hongseok Namkoong, John Miller, Hannaneh Hajishirzi, Ali Farhadi, and Ludwig Schmidt. Openclip, July 2021. URL https://doi.org/10.5281/zenodo.5143773

    2021 doi

  26. [34]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In CVPR, 2019

    2019

  27. [35]

    Training generative adversarial networks with limited data

    Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training generative adversarial networks with limited data. In NeurIPS, 2020 a

    2020

  28. [36]

    Analyzing and improving the image quality of stylegan

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of stylegan. In CVPR, 2020 b

    2020

  29. [37]

    Alias-free generative adversarial networks

    Tero Karras, Miika Aittala, Samuli Laine, Erik H \"a rk \"o nen, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Alias-free generative adversarial networks. In NeurIPS, 2021

    2021

  30. [38]

    Imagic: Text-based real image editing with diffusion models

    Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. Imagic: Text-based real image editing with diffusion models. In CVPR, 2023

    2023

  31. [39]

    Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on

    Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. In CVPR, 2024

    2024

  32. [40]

    Auto-encoding variational bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes. In ICLR, 2014

    2014

  33. [41]

    High-resolution virtual try-on with misalignment and occlusion-handled conditions

    Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In ECCV, 2022

    2022

  34. [42]

    Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing

    Dongxu Li, Junnan Li, and Steven Hoi. Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing. In NeurIPS, 2023 a

    2023

  35. [43]

    Toward accurate and realistic outfits visualization with attention to details

    Kedan Li, Min Jin Chong, Jeffrey Zhang, and Jingen Liu. Toward accurate and realistic outfits visualization with attention to details. In CVPR, 2021

    2021

  36. [44]

    Warpdiffusion: Efficient diffusion model for high-fidelity virtual try-on

    Xiu Li, Michael Kampffmeyer, Xin Dong, Zhenyu Xie, Feida Zhu, Haoye Dong, Xiaodan Liang, et al. Warpdiffusion: Efficient diffusion model for high-fidelity virtual try-on. arXiv preprint arXiv:2312.03667, 2023 b

    2023 arXiv

  37. [45]

    One-stage 3d whole-body mesh recovery with component aware transformer

    Jing Lin, Ailing Zeng, Haoqian Wang, Lei Zhang, and Yu Li. One-stage 3d whole-body mesh recovery with component aware transformer. In CVPR, 2023

    2023

  38. [46]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In ICLR, 2019

    2019

  39. [47]

    Dress code: High-resolution multi-category virtual try-on

    Davide Morelli, Matteo Fincato, Marcella Cornia, Federico Landi, Fabio Cesari, and Rita Cucchiara. Dress code: High-resolution multi-category virtual try-on. In CVPR, 2022

    2022

  40. [48]

    Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on

    Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcella Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In ACM MM, 2023

    2023

  41. [49]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models

    Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In AAAI, 2024

    2024

  42. [50]

    Dreammatcher: Appearance matching self-attention for semantically-consistent text-to-image personalization

    Jisu Nam, Heesu Kim, DongJae Lee, Siyoon Jin, Seungryong Kim, and Seunggyu Chang. Dreammatcher: Appearance matching self-attention for semantically-consistent text-to-image personalization. In CVPR, 2024

    2024

  43. [51]

    To create what you tell: Generating videos from captions

    Yingwei Pan, Zhaofan Qiu, Ting Yao, Houqiang Li, and Tao Mei. To create what you tell: Generating videos from captions. In ACM MM, 2017

    2017

  44. [52]

    Boosting diffusion models with moving average sampling in frequency domain

    Yurui Qian, Qi Cai, Yingwei Pan, Yehao Li, Ting Yao, Qibin Sun, and Tao Mei. Boosting diffusion models with moving average sampling in frequency domain. In CVPR, 2024

    2024

  45. [53]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, 2021

    2021

  46. [54]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj \"o rn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, 2022

    2022

  47. [55]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015

    2015

  48. [56]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In CVPR, 2023

    2023

  49. [57]

    Very deep convolutional networks for large-scale image recognition

    K Simonyan and A Zisserman. Very deep convolutional networks for large-scale image recognition. In ICLR, 2015

    2015

  50. [58]

    Objectstitch: Object compositing with diffusion model

    Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, and Daniel Aliaga. Objectstitch: Object compositing with diffusion model. In CVPR, 2023

    2023

  51. [59]

    Loftr: Detector-free local feature matching with transformers

    Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, and Xiaowei Zhou. Loftr: Detector-free local feature matching with transformers. In CVPR, 2021

    2021

  52. [60]

    Emergent correspondence from image diffusion

    Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, and Bharath Hariharan. Emergent correspondence from image diffusion. In NeurIPS, 2023

    2023

  53. [61]

    Improving virtual try-on with garment-focused diffusion models

    Siqi Wan, Yehao Li, Jingwen Chen, Yingwei Pan, Ting Yao, Yang Cao, and Tao Mei. Improving virtual try-on with garment-focused diffusion models. In ECCV, 2024

    2024

  54. [62]

    Toward characteristic-preserving image-based virtual try-on network

    Bochao Wang, Huabin Zheng, Xiaodan Liang, Yimin Chen, Liang Lin, and Meng Yang. Toward characteristic-preserving image-based virtual try-on network. In ECCV, 2018

    2018

  55. [63]

    Tracking everything everywhere all at once

    Qianqian Wang, Yen-Yu Chang, Ruojin Cai, Zhengqi Li, Bharath Hariharan, Aleksander Holynski, and Noah Snavely. Tracking everything everywhere all at once. In ICCV, 2023

    2023

  56. [64]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE TIP, 2004

    2004

  57. [65]

    Gp-vton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning

    Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. Gp-vton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning. In CVPR, 2023

    2023

  58. [66]

    Ootdiffusion: Outfitting fusion based latent diffusion for controllable virtual try-on

    Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Ootdiffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024 a

    2024 arXiv

  59. [67]

    Magicanimate: Temporally consistent human image animation using diffusion model

    Zhongcong Xu, Jianfeng Zhang, Jun Hao Liew, Hanshu Yan, Jia-Wei Liu, Chenxu Zhang, Jiashi Feng, and Mike Zheng Shou. Magicanimate: Temporally consistent human image animation using diffusion model. In CVPR, 2024 b

    2024

  60. [68]

    Hi3d: Pursuing high-resolution image-to-3d generation with video diffusion models

    Haibo Yang, Yang Chen, Yingwei Pan, Ting Yao, Zhineng Chen, Chong-Wah Ngo, and Tao Mei. Hi3d: Pursuing high-resolution image-to-3d generation with video diffusion models. In ACM MM, 2024

    2024

  61. [69]

    Inpaint anything: Segment anything meets image inpainting

    Tao Yu, Runseng Feng, Ruoyu Feng, Jinming Liu, Xin Jin, Wenjun Zeng, and Zhibo Chen. Inpaint anything: Segment anything meets image inpainting. arXiv preprint arXiv:2304.06790, 2023

    2023 arXiv

  62. [70]

    Pymaf-x: Towards well-aligned full-body model regression from monocular images

    Hongwen Zhang, Yating Tian, Yuxiang Zhang, Mengcheng Li, Liang An, Zhenan Sun, and Yebin Liu. Pymaf-x: Towards well-aligned full-body model regression from monocular images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 0 (10): 0 12287--12303, 2023 a

    2023

  63. [71]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, 2023 b

    2023

  64. [72]

    Trip: Temporal residual learning with image noise prior for image-to-video diffusion models

    Zhongwei Zhang, Fuchen Long, Yingwei Pan, Zhaofan Qiu, Ting Yao, Yang Cao, and Tao Mei. Trip: Temporal residual learning with image noise prior for image-to-video diffusion models. In CVPR, 2024

    2024

  65. [73]

    Unipc: A unified predictor-corrector framework for fast sampling of diffusion models

    Wenliang Zhao, Lujia Bai, Yongming Rao, Jie Zhou, and Jiwen Lu. Unipc: A unified predictor-corrector framework for fast sampling of diffusion models. In NeurIPS, 2023

    2023

  66. [74]

    Tryondiffusion: A tale of two unets

    Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Saharia, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In CVPR, 2023

    2023

  67. [75]

    Sd-dit: Unleashing the power of self-supervised discrimination in diffusion transformer

    Rui Zhu, Yingwei Pan, Yehao Li, Ting Yao, Zhenglong Sun, Tao Mei, and Chang Wen Chen. Sd-dit: Unleashing the power of self-supervised discrimination in diffusion transformer. In CVPR, 2024

    2024

  68. [76]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  69. [77]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  70. [78]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  71. [79]

    The task empowers the end users to experience the visual affects of wearing various clothings without the need of physical store try-ons

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.